At Snowflake World Tour London, Snowflake Chief Security, Trust Officer Mayank Uphadyay delivered a clear message: enterprises racing to deploy AI agents cannot afford to skip the trust and governance layer. Uphadyay argued that security has shifted from a blocker to an enabler of AI adoption, and companies must build confidence in their data foundations before agents become autonomous.

What You Need to Know

Uphadyay highlighted a two-sided security challenge: preventing external AI attacks on the platform while ensuring internal AI agents do not leak data or break boundaries. Snowflake's Trust Center is the key tool that connects data, models and agents with proper guardrails. The company advocates a "rock solid" data foundation where permissions are carefully scoped for each agent. As enterprises move from experiments to production, governance frameworks become the difference between safe automation and costly breaches.

The Governance Challenge for AI Agents

Uphadyay described a spectrum of AI readiness among Snowflake customers. The central problem, he said, is that agents are exploratory by nature and can accumulate excessive permissions over time, similar to human employees who retain access to systems they no longer use. Ensuring agents operate with least-privilege permissions is critical to avoiding data exfiltration or accidental system breaches.

  • Agent identity: Each AI agent must have a clear identity and permission set that is regularly audited.
  • Perimeter control: The platform must defend against malicious AI targeting the data layer.
  • Behavioral guardrails: Agents must be prevented from taking actions outside their designated scope.

"When customers trust us with their data, we must earn that trust every day," Uphadyay said. He likened the security architecture to a race car: brakes enable speed because they provide confidence. "Security is what helps you build that trust," he added.

Why This Matters

The rapid rise of autonomous AI agents marks a turning point for enterprise security. Traditional perimeter-based defenses are inadequate when agents can act on behalf of users and access multiple systems. Snowflake's approach, centered on the Trust Center, offers a model for how platforms can embed governance without slowing innovation. For businesses, the implication is clear: those that fail to implement proper agent identity, permission management and monitoring risk not only data loss but also erosion of customer trust. This shift from security as a barrier to security as an enabler changes the entire adoption equation for AI tools across industries.

Uphadyay on the Path Forward

Uphadyay described a pyramid of automation where mundane tasks are handled by AI first, then progressively more critical work as confidence grows. Human oversight remains essential at the top of the pyramid. "Things are here, things are real," he said. "It is here and now." This perspective reinforces that the agentic enterprise is not a future concept but a present reality requiring immediate governance attention. Snowflake's Trust Center runs continuously across the data, model and agent layers, providing visibility into what each agent does and what improvements are needed.

Ensuring the right permissions for each agent is an ongoing process. Uphadyay noted that Snowflake's 12-year track record of data security has built the trust foundation that makes agentic adoption possible. The company already offers tools like Coco and CoWork to help customers automate tasks while maintaining control. The challenge now is scaling these practices as agents become more autonomous and interconnected.